Improved Detection of Early Diabetic Retinopathy using Contrast Limited Adaptive Histogram Equalization and Principal Component Analysis with Bag of Visual Words based SVM

Diabetes is a serious and complex disease that impacts numerous vital organs within the human body, with the eyes being particularly vulnerable to its effects. The proposed method for detecting diabetic retinopathy involves preprocessing the data to ensure coherence and focus on relevant features. The dataset consists of 130 color fundus images, with 20 normal and 110 exhibiting signs of diabetic retinopathy. The preprocessing process involves color space conversion, image restoration, and enhancement stages. Techniques like the Median Filter and Contrast Limited Adaptive Histogram Equalization (CLAHE) are used to reduce noise and improve image quality. The system also employs localization and segmentation techniques to pinpoint and isolate the optic disc within the images. The proposed system uses a Bag Of Visual Words (BOW) based SVM classification techniques to categorize the disease's severity levels, contributing to improved diagnosis and treatment planning in diabetic retinopathy cases. The data reveals an impressive overall accuracy rate of 99.006%, affirming the system's reliability and practical utility. These results underscore the system's potential to provide valuable support to ophthalmologists in image classification, thereby serving as a tool to assist in informed decision-making processes within the field.